Florida Attorney General James Uthmeier says his office is investigating OpenAI over alleged links between ChatGPT and the 2025 Florida State University shooting. The state’s public case, though, already reaches beyond that attack into child safety, self-harm, and data privacy concerns.
The public record of the FSU case remains incomplete. Attorneys for a victim’s family say the accused gunman used ChatGPT before the shooting, and prosecutors have confirmed relevant records exist. The chatbot’s actual replies have not been made public.
Probe stretches past one attack
According to TechCrunch’s report on the investigation, Uthmeier announced on April 9 that subpoenas are forthcoming as part of a formal OpenAI probe tied to the FSU shooting. TechCrunch reported that the April 2025 attack killed two people and injured five, and that attorneys for a victim’s family claim ChatGPT was used to help plan it.
The investigation is not limited to that one incident. Bloomberg Law reported on Florida’s probe that Uthmeier also cited allegations involving child sexual abuse material, self-harm, and concerns that user data could be exposed to foreign adversaries. Bloomberg Law also reported that the subpoenas were not yet public and that the matter is being handled through Florida’s state data privacy unit, suggesting Florida is pursuing it through consumer-protection authority rather than criminal law.
OpenAI said it would cooperate. TechCrunch reported that the company said more than 900 million people use ChatGPT each week and that it builds the product to understand intent and respond in a safe and appropriate way. WPTV’s coverage of the case also said that after learning about the shooting in April 2025, OpenAI identified an account believed to be associated with the suspect and proactively shared that information with law enforcement.
Missing replies remain the key evidence gap
WPTV reported that records tied to accused gunman Phoenix Ikner include queries about when the FSU Student Union would be busiest, how the country would react to a campus shooting, how many victims typically draw media attention, and questions related to suicide. But the available reporting does not include ChatGPT’s full replies. Without those responses, it is harder to tell whether the logs show operational assistance, attempted refusal, or primarily a record of the suspect’s state of mind.
One day before the Florida announcement, OpenAI published its Child Safety Blueprint, a policy framework calling for stronger laws, improved reporting, and safety-by-design measures aimed at AI-enabled child exploitation. That document does not answer Florida’s allegations, but it does show OpenAI was publicly arguing for tougher safety standards as the state moved to investigate it.